Tensorflow.js Error: Backend name ‘webgpu’ not found in registry when not served on localhost

Question

I am working on getting a local server set up that can serve pages that will have access to tensorflow.js and the webGPU backend.

I have run into an issue where if I serve my files on localhost (127.0.0.1) I am able to load the webGPU backend and everything runs as expected.

However, when I try serving the exact same files at my local IP address like 192.168.x.x (so I can access the host from other devices), I get a “Error: Backend name ‘webgpu’ not found in registry” error when trying to set the backend to webGPU.

I tried serving it using purely node with the http module, http-server, and express and am running into the same error.

Anyone have any ideas of what could cause this to happen? Are there specific headers that need to be set in order to support the webGPU backend? What could cause this to break when all I have done is change the IP address where the file is served

await import(`https://cdn.jsdelivr.net/npm/@tensorflow/[email protected]/dist/tf.min.js`);
await import(`https://cdn.jsdelivr.net/npm/@tensorflow/tfjs-backend-webgpu/dist/tf-backend-webgpu.js`);
await tf.ready();
await tf.setBackend("webgpu");   // <--Error: Backend name 'webgpu' not found in registry
console.log(tf.getBackend());

François Beaufort · Accepted Answer · 2024-03-11 09:34:34Z

6

I believe it happens because you're not serving your files over HTTPS. See https://developer.chrome.com/docs/web-platform/webgpu/troubleshooting-tips#the_gpu_adapter_is_null

WebGPU is accessible only to secure contexts. If you serve your code over an insecure protocol (for example, http:, file:), either use the secure https: protocol or address this during the development of your web app in one of the following ways:

Serve your code locally on http://localhost or http://127.0.0.1 with either of these commands: npx http-server or python3 -m http.server.

Add the origin to the "Insecure origins treated as secure" list of chrome://flags/#unsafely-treat-insecure-origin-as-secure and restart Chrome.

Install Node.js and run npx servez --ssl to serve your folder over https with a fake certificate. You'll still get a warning in Chrome you can bypass by clicking "Advanced" then "Proceed to...".

Expose your local web server to the Internet with ngrok.

answered Mar 11, 2024 at 9:34

François Beaufort

5,8093 gold badges31 silver badges42 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Occam Over a year ago

Yes, this seems to be the problem. What a confusing issue to try and troubleshoot. You would think the browser would issue some warning to point you in the right direction

Collectives™ on Stack Overflow

Tensorflow.js Error: Backend name ‘webgpu’ not found in registry when not served on localhost

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related